Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Sentence Representation via Contrastive Learning with Mixing Negatives
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add mixed hard negatives by mixing positive and randomly sampled negative features | SimCSE-BERTbase · STS12-16, STS-B, SICK-R (Avg) | Spearman correlation: 74.83±2.32 -> 77.66±0.61 | Helped |
| add mixed hard negatives by mixing positive and randomly sampled negative features | SimCSE-BERTlarge · STS12-16, STS-B, SICK-R (Avg) | Spearman correlation: 77.97±0.7 -> 78.80±0.09 | Helped |
| add mixed hard negatives by mixing positive and randomly sampled negative features | SimCSE-BERTbase · Transfer tasks (Avg) | classification accuracy: 84.49±0.59 -> 86.33±0.26 | Helped |
| add mixed hard negatives by mixing positive and randomly sampled negative features | SimCSE-BERTlarge · Transfer tasks (Avg) | classification accuracy: 86.8±0.42 -> 87.77±0.11 | Helped |
| add mixed negatives for both anchor views rather than only one (ablation) | MixCSE-BERTbase · STS (Avg) | Spearman correlation: 76.43±0.11 -> 77.66±0.61 | Helped |
| add mixed negatives for both anchor views rather than only one (ablation) | MixCSE-BERTbase · Transfer tasks (Avg) | classification accuracy: 85.86±0.04 -> 86.33±0.26 | Helped |
| add mixed negatives for both anchor views rather than only one (ablation) | MixCSE-BERTlarge · STS (Avg) | Spearman correlation: 78.11±0.20 -> 78.80±0.09 | Helped |
| add mixed negatives for both anchor views rather than only one (ablation) | MixCSE-BERTlarge · Transfer tasks (Avg) | classification accuracy: 87.21±0.23 -> 87.77±0.11 | Helped |
| apply stop-gradient to mixed negatives (ablation) | MixCSE-BERTbase · STS (Avg) | Spearman correlation: 75.74±0.87 -> 77.66±0.61 | Helped |
| apply stop-gradient to mixed negatives (ablation) | MixCSE-BERTbase · Transfer tasks (Avg) | classification accuracy: 84.55 ± 0.27 -> 86.33±0.26 | Helped |
| apply stop-gradient to mixed negatives (ablation) | MixCSE-BERTlarge · STS (Avg) | Spearman correlation: 77.61±0.45 -> 78.80±0.09 | Helped |
| apply stop-gradient to mixed negatives (ablation) | MixCSE-BERTlarge · Transfer tasks (Avg) | classification accuracy: 86.02±0.27 -> 87.77±0.11 | Helped |
Closest papers by meaning
- Instance Smoothed Contrastive Learning for Unsupervised Sentence Embedding (AAAI 2023 · similarity 0.89)
- TNCSE: Tensor Norm Constraints for Unsupervised Contrastive Learning of Sentence Embeddings (AAAI 2025 · similarity 0.88)
- Narrowing the Gap between Supervised and Unsupervised Sentence Representation Learning with Large Language Model (AAAI 2024 · similarity 0.88)
- DenoSent: A Denoising Objective for Self-Supervised Sentence Representation Learning (AAAI 2024 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.86)
- Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.